The Credibility Revolution

Regression Discontinuity Designs in Economics

Properties1
tagsliterature

2 min read

Regression Discontinuity Designs in Economics

Causal Question / Estimand

The treatment effect at the cutoff — a local [[Causal-Estimand|average treatment effect]] (a LATE in the fuzzy case) — when treatment is determined by whether a running variable crosses a known threshold.

Identification Strategy

A “user’s guide” to RD. Identification rests on Continuity-at-Cutoff: the conditional expectation of potential outcomes is continuous in the running variable at the threshold, so the jump in observed outcomes at the cutoff is causal. Lee’s key behavioural argument: when units cannot precisely manipulate the running variable (No-Manipulation), assignment near the cutoff is “as good as randomized,” giving RD the internal validity of an experiment locally. Distinguishes sharp RD (treatment is a deterministic step at the cutoff) from fuzzy RD (the cutoff shifts treatment probability → an IV/LATE estimand). Recommends local linear regression, graphical analysis, and treating bandwidth/specification as central choices.

Key Assumptions

Continuity-at-Cutoff (continuity of potential-outcome expectations), No-Manipulation (no precise sorting), and — for fuzzy RD — the IV triad (Exclusion-Restriction, Monotonicity) at the cutoff. SUTVA.

Threats to Validity

Manipulation/sorting (test via McCrary density and covariate continuity); sensitivity to bandwidth and polynomial order; extrapolation away from the cutoff (external validity is local). Recommends placebo cutoffs and covariate-balance checks.

Setting / Data

n/a — methodological survey; draws on canonical RD applications (elections, class size, education thresholds).

Key Claims

  • RD has strong internal validity because, under imprecise manipulation, the cutoff generates locally randomized variation.
  • Estimates are inherently local to the cutoff — a feature, not a bug, but it limits external validity.
  • Use local linear regression, graph the discontinuity, and report sensitivity to bandwidth; test for manipulation and covariate jumps.

Connections

Citation

Lee, D. S., & Lemieux, T. (2010). Regression Discontinuity Designs in Economics. Journal of Economic Literature, 48(2), 281–355.

Graph View

Backlinks